Mem graph doesn't update after crash of the program when used all memory
martincerven opened this issue · comments
Martin Cerven commented
Describe the bug
When using models that consume all available memory and eventually crash, the jtop doesn't update, only after ctrl+c, jtop it shows available memory correctly.
To Reproduce
Steps to reproduce the behavior:
- use audiocraft on Orin Nano with monitor
- When running jupyter cells, it will eventually run out of memory and crash container
- Mem won't update, only after closing and reopening
Expected behavior
I would expect I won't have to restart jtop
Additional context
When running any (pytorch) models that consume more memory than available, the jtop doesn't update after they eventually crash.
Board
Happened to both Orin Nano 8GB devkit, when trying audiocraft, and AGX 32GB devkit when running larger models (in containers)